when mpii(4) gets an async event (ie, sas topology change or

integrated raid has changed state) that needs an acknowledgement,
it currently goes scsi_io_get. because events are handled at interrupt
time it cannot wait for ccbs, so if the ccb pool is empty, it wont
get a ccb to ack the event with.

this cuts the code over to using iohandlers to ack these events.
when an ack is required, it shoves the rcb onto a queue and schedules
an iohandler to be run whenever a ccb becomes available. the majority
of the time the ccb will be there and the reply will happen
immediately. however, if the hba is busy we might be deferred until
some other command completes and gives its ccb back to the pool.

because it looks like we can now receive many events at a time
(rather than just one at a time like in mpi), im using a queue to
store the events in. the alternative was an iohandler per rcb, but
that struck me as a bit too heavy.

mikeb is really impressed
no objection from krw@

0 files changed, 0 insertions, 0 deletions